# Thread: Coefficients in linear regression

1. ## Coefficients in linear regression

hi i am using spss 17.0 i am using linear regression to predict behaviours

for the coefficent output does the constant and the indperpentant (both under the model bit) have to be signifcant to continue to use the equation for predicting behaviour?
the k-s test in non-sig for the Unstandardized Residual data. hope this is a enough info to answer the question, thanks

2. ## Re: Coefficients in linear regression

The constant need not be but independent variable need to be

3. ## Re: Coefficients in linear regression

Well there's a little bit more to it than that. You need to make sure the model has an adequate fit and you really don't want to extrapolate outside of your dataset unless you have theoretical reasons why your data should have an expectation function like you specified.

4. ## Re: Coefficients in linear regression

Keep in mind that the significant linear predictor may not necessarily tell you if the regression is good for prediction. That simply will tell you that there's a relationship. Even the standard error of the parameter estimate won't be a good indicator of prediction, since this tells you the distribution of the mean y at a given x. Good for population-level prediction, but not for individual prediction, as it doesn't take into account the variance attributable to error in your model.

To that, I would add a suggestion to ask yourself how you're using the the regression for prediction. I've been dealing with the issue of prediction in a Generalized Additive Mixed Model a bit, and it's a real pain. Simpler for linear regression, so that's good for you. I would think about using prediction intervals instead of confidence intervals if you want to predict y from a single value of x. If you want to understand the predicted mean (within your range of values), you can use the confidence interval. Others can add to my horrible explanation of this. Please chastise me for unsolicited advice. I just don't want anyone to go through the monumental days of frustration that I just went through.

5. ## Re: Coefficients in linear regression

Originally Posted by jpkelley
Keep in mind that the significant linear predictor may not necessarily tell you if the regression is good for prediction. That simply will tell you that there's a relationship. Even the standard error of the parameter estimate won't be a good indicator of prediction, since this tells you the distribution of the mean y at a given x. Good for population-level prediction, but not for individual prediction, as it doesn't take into account the variance attributable to error in your model.

To that, I would add a suggestion to ask yourself how you're using the the regression for prediction. I've been dealing with the issue of prediction in a Generalized Additive Mixed Model a bit, and it's a real pain. Simpler for linear regression, so that's good for you. I would think about using prediction intervals instead of confidence intervals if you want to predict y from a single value of x. If you want to understand the predicted mean (within your range of values), you can use the confidence interval. Others can add to my horrible explanation of this. Please chastise me for unsolicited advice. I just don't want anyone to go through the monumental days of frustration that I just went through.
i sent a pm have no idea if it went through?

6. ## Re: Coefficients in linear regression

What did your PM say? Feel free to ask any questions in this thread so that others can provide input as well.

7. ## Re: Coefficients in linear regression

Originally Posted by Dason
What did your PM say? Feel free to ask any questions in this thread so that others can provide input as well.
lol firstly i dont know much about stats so i get lost easily with the terms used like in these replys

i have just realised that i want to predict amount of exercise done per week (mins) which i put as a dependant variable and have changed the y= mc*X to X=y = M/C to get a prediction i did this seperately for 2 varibles and now going on to mutiple regression then i realised that exercise done per week (mins) should be inderpendant but when i do this it is non-sig and non parametric (the rest of the variables are parametric
question 1) is this loophole if u like allowed to be used in linear regression so that i can predict exericse mins by rearraging the eqaution
2) what would the soultion be so that i can use mutiplte regression? (tranforming the data still comes up non-sig) do i have use nonparametric instead? considering the rest of my thesis i have used parametric tests (not using exercise per week (mins) in any of them) and the final hypo is predicting behaviours.
thanks, hope that makes sensee

 Tweet

#### Posting Permissions

• You may not post new threads
• You may not post replies
• You may not post attachments
• You may not edit your posts